So a prompt engineer might be employed by a company
when communicating with ChatGPT and Perplexity?
there is no evidence that that improves the results,
but being emotional in your prompts, for example,
using all caps does actually improve performance as well.
that when someone says this is very important for my career,
that you should work a little bit less in the holidays.
So I've asked it to tell me about quantum mechanics
And now it's explaining quantum mechanics as a magic world
where tiny things like atoms can be in two places at once.
As an astrophysicist or you are an astrophysicist,
One person who is famous at product naming was Steve Jobs.
of naming companies, you can provide some examples
if you put titles on the different sections of your prompt.
@petemandik is asking serious question about AI artists,
and the physics is quite difficult to understand
so they hadn't really learned how the world works yet.
We also have a really strong eye
for whether fingers are wrong or whether eyes are wrong.
because everything in a prompt is positively weighted.
If you say don't put a picture of an elephant in the room,
then it will actually introduce a picture of an elephant.
For example, it's not currently available in Dall-E,
but it is available in Stable Diffusion.
So we're gonna type in oil painting hanging in a gallery.
One thing I can do is if I add up the negative prompt here
what is the weirdest response you've gotten from ChatGPT?
it doesn't know that Mary Lee Pfeiffer's son is Tom Cruise,
but there's not that much information on the internet
about Mary Lee Pfeiffer and who her son is.
Hallucinating is when the AI makes something up that's wrong
and it's really hard to get away from hallucination
creativity is really just hallucinating something
then won't that bias come through in its responses?
Well you're absolutely correct
because AI are trained on all of the data from the internet
and the internet is full of bias
because it comes from us and humans are biased too,
but it can be pretty hard to correct for those bias
by adding guardrails
because by trying to remove bias in one direction,
you might be adding bias in another direction.
Famous example was when Google added to their prompts
for their AI image generator service,
an instruction that they should always show diverse people
and it would never create a white George Washington.
In trying to do the right thing and solve for one bias,
There is a lot of work in the research labs like Anthropic
as a whole safety research team that have figured out,
@carkujon wants to know how much of the conversation context
does ChatGPT actually remember?
If we chatted for a year with information dense messages,
would it be able to refer back to info from a year ago?
unless you put something in your settings specifically.
is I will get all of the context at one thread for a task.
I'll just ask it to summarize
and then I'll take that summary and then start a new thread
and then I've got the summary more condensed information.
@bybrandonwhite, does customizing your settings in ChatGPT
and providing your bio/personal info help better results?
Yes, I find that you get wildly different results
when you put some information in the custom instructions.
You have two fields.
Custom instructions, which is what would you like ChatGPT
that you get annoyed about when you're using ChatGPT,
A prompt engineer is designing that system of prompts
that a civil engineer is doing with a bridge, right?
Like, they're designing the bridge
and they're making sure that when you drive over it,
that work in managing humans also work in managing AIs.
in order to predict the next token or word in the sentence.
and we have Lakers, which has a 13% chance of coming next.
We also have the word Los,
which is just the beginning of the word Los Angeles.
It's not always picking the highest probability word,
and it can be trained to be used in different contexts .
@AdeyemiTestimo4, what is the best LLM in your opinion?
of five product names for a shoe that fits any foot size.
We're testing the model's creativity here,
and you can see that we have UniFit Shoes as one idea.
And you can see it comes up with really different names,
For me personally, it's been the programming ability
that I get from using ChatGPT and Anthropic's Claude.
I could just copy that and then paste it at the bottom
And that's the way that you learn with programming.
I find that I just never get stuck when I use this.
One of the coolest things, I've done a little automation
you get much better results and they're comprehensive
Not only does it make the thought process more observable
because you can see what happened at each step
and which steps failed, but also the LLM gets less confused
because you don't have a huge prompt
with lots of different conflicting instructions
where it's running in a loop and it keeps prompting itself
Microsoft AutoGen is a framework for autonomous agents
that you can just say, make me more money for the company
how would you prompt the LLM to improve the prompt?
techniques like the automatic prompt engineer technique
I actually use it all of the time to optimize my prompts.
Just because you're a prompt engineer
doesn't necessarily mean you're immune from your work
being automated as well.
@BahouPrompts is asking how long until prompt engineering
or a future field related to this becomes a degree?
because of course humans are already pretty intelligent
and we need prompting.
We have a HR team, we have a legal team, we have management.
So I think that the practice
of prompt engineering will always be a skill that you need
Thanks for watching Prompt Engineering Support.
